Papers for APTS Advisory Committee Meeting AC6
13th September, 2012
• Report for item 2.2.3: APTS Alumni Questionnaire Summary (2007/08 Cohort)
• Report for item 3: APTS 2011/12
• Registration information for 2012/13
• Billing and Cancellation policy
• Paper from Tim Davis for item 4.2.
• Paper from Jonathan Rougier for item 4.3.
• Current Statistical Inference syllabus for item 4.3.
• Written contributions received from:
– David Elston (BioSS)
– Jonathan Rougier (Bristol)
– Dave Woods (Southampton)
– Shuaiwei Zhou (Student Representative; TCD)
1
Alumni Questionnaire: Preliminary Summary
Alumni of the 2007/08 APTS weeks were contacted via email where possible and via academic contacts
at sending institutions in some others instances and asked to complete an online questionnaire (available
at: http://www2.warwick.ac.uk/fac/sci/statistics/apts/alumniquestionnaire )
Thus far we have had 28 responses which are summarised on the following page.
Table Notes
Those who answered “other” to question 1 provided the following details:
• Lecturer in statistics in a University
• Looking for a position
• Research fellow
• Studying for a part time PhD and working as a statistical scientist for the European Commission
The “other” response to question 6 was accompanied by “A great opportunity to be taught at post
grad level rather than reading books - which always takes longer for things to sink in.”
1
2
8. Would you recommend APTS to someone just starting a PhD in applied
probability or statistics?
7. Which of the APTS modules ended up connected closely to your current
employment? (specify as many as is appropriate, answer only if you are not still
working for your PhD)
6. If your answer to either of the previous two questions was yes, which of the
following reasons underlie your recommendation? (specify as many as is appropriate)
4. Has the training given by APTS proved helpful to you in your PhD experience
and research?
5. Has the training given by APTS proved helpful to you in your present
employment? (answer only if you are not still working for your PhD)
3. Which of the APTS modules ended up connected closely to your eventual PhD
research? (specify as many as is appropriate)
2. Which of the APTS weeks did you attend in 2007-08? (specify as many as is
appropriate)
1. What best describes your current status?
Question
Simple summary
Response
still studying for PhD
working in industry as statistical scientist
working in industry in another role
working in academia as statistical scientist
working in academia in another role
other (please specify)
Statistical Computing & Statistical Inference
Statistical Modelling & Statistical Asymptotics
Applied Stochastic Processes & Computer Intensive Statistics
Spatial and Longitudinal Data Analysis & Nonparametric
Smoothing
Statistical Computing
Statistical Inference
Statistical Modelling
Statistical Asymptotics
Applied Stochastic Processes
Computer Intensive Statistics
Spatial and Longitudinal Data Analysis
Nonparametric Smoothing
Yes
No
Yes
No
The modules provide a broad general training in the area
APTS weeks enable networking with peers
APTS weeks enable broader contacts with more senior academics
Other (please specify)
Statistical Computing
Statistical Inference
Statistical Modelling
Statistical Asymptotics
Applied Stochastic Processes
Computer Intensive Statistics
Spatial and Longitudinal Data Analysis
Nonparametric Smoothing
Yes
No
12
10
13
3
5
13
6
7
26
2
19
3
24
20
9
1
9
12
16
5
5
11
8
7
26
2
16
1
3
20
23
21
20
Number
6
2
Paper for 3. APTS year 5 (2011–12):
Summary report to Advisory Committee
Preliminary: some summary counts for all five years 2007–2012
Numbers of students who took at least one APTS week (of which, EPSRC-funded in brackets):
2007-2008
88 (38)
2008-2009
88 (37)
2009-2010
100 (45)
2010-2011
90 (37)
2011-2012
125 (46)
Number of APTS lecturers to date: 15. (This will increase to 17 in 2012–13).
Number of APTS-week host institutions to date: 9.
Member Institutions
In 2011-12 there were 29 MIs, all located in the UK and Ireland.
APTS weeks, academic year 2011–12
Week 1, January 2011, Cambridge:
• Statistical Computing (S N Wood)
• Statistical Inference (D Firth)
• Evening sessions:
– RSS Reception
– Pub Quiz
Week 2, April 2012, Nottingham:
• Statistical Modelling (J J Forster & D Woods)
• Statistical Asymptotics (A Wood)
• Evening sessions:
– RSS Reception
Week 3, July 2012, Warwick
• Applied Stochastic Processes (S B Connor and C A Goldschmidt)
• Computer Intensive Statistics (B D Ripley)
• Special lecture: Prof. Sir. David Cox
• Evening sessions:
– RSS Reception
– Statistics at Large a presentation by statisticians
– Trip to Kennilworth pub trip organised by PhD students
Week 4, September 2012, Glasgow:
• Spatial and Longitudinal Data Analysis (P J Diggle)
• Nonparametric Smoothing (R J Samworth)
• Evening sessions:
1
– RSS Reception
– Pub Quiz
– Ceilidh
Registrations
A total of 128 unique students were registered to attend one or more APTS week. Numbers of registrations for each of the four weeks were 96, 81, 96 and 84, respectively whilst corresponding attendance
figures were 92, 74, 85 and 66, respectively. This year we were able to accommodate all registered
students.
Of those 129 students:
• 56 were registered to attend all 4 weeks and 40 students actually did so.
• 121 were first year students in statistics or probability, of whom 47 were EPSRC funded.
• APTS Member Institutions supplied 112 of the 128 applications.
Student feedback
The following summarizes student responses to an anonymous questionnaire completed at the end of
each training week.
A. Preparation for APTS
1. Did you find the APTS web site useful?
Week
Week
Week
Week
1
2
3
4
Yes
81
62
43
31
No
0
0
1
0
Didn’t use
3
0
0
1
2. Was it clear exactly what was expected of you in the weeks leading up to this APTS week? (e.g.,
making travel arrangements, looking at preliminary material, registering your meal choice, etc.)
Week
Week
Week
Week
1
2
3
4
Yes
83
62
44
32
2
No
1
0
0
2
3. Roughly, how many days did you spend on the preliminary material?
Histogram of the number of days spent on the preliminary material of
Statistical Inference
20
Frequency
20
15
0
0
5
10
10
Frequency
25
30
30
40
35
Histogram of the number of days spent on the preliminary material of
Statistical Computing
0
2
4
6
8
10
12
14
0
2
4
6
8
10
12
Number of days
Histogram of the number of days spent on the preliminary material of
Statistical Modelling
Histogram of the number of days spent on the preliminary material of
Statistical Asymptotics
14
20
Frequency
15
15
0
0
5
5
10
10
Frequency
20
25
30
25
35
30
Number of days
2
4
6
8
10
2
6
8
10
Histogram of the number of days spent on the preliminary material of
Applied Stochastic Processses
10
0
0
5
10
Frequency
15
Histogram of the number of days spent on the preliminary material of
Computer Intesive Statistics
15
Number of days
5
Frequency
4
Number of days
2
4
6
8
10
0
2
4
6
8
Number of days
Number of days
Histogram of the number of days spent on the preliminary material of
Nonparametric Smoothing
Histogram of the number of days spent on the preliminary material of
Spatial and Longitudinal Data Analysis
10
10
Frequency
0
0
5
5
Frequency
10
15
15
0
0
2
4
6
8
10
0
Number of days
2
4
6
Number of days
4. Did the preliminary material help you to understand the lectures this week?
3
8
10
Yes
75
78
40
48
32
28
12
20
Statistical Computing
Statistical Inference
Statistical Asymptotics
Statistical Modelling
Stochastic Processes
Computer Intensive
Nonparametric Smoothing
Spatial and Longitudinal
No
3
1
8
2
1
4
6
2
Didn’t use
6
5
14
12
10
11
17
13
B. The APTS week: material covered
1. How would you rate the level of the module lectures?
Statistical Computing
Statistical Inference
Statistical Asymptotics
Statistical Modelling
Stochastic Processes
Computer Intensive
Nonparametric Smoothing
Spatial and Longitudinal
Too easy
3
6
1
1
1
2
0
0
Just right
67
67
23
58
35
37
20
30
Too hard
11
7
35
1
8
5
14
4
2. Did the Oxford R lectures help you with the computer sessions during the week?
Week
Week
Week
Week
1
2
3
4
Yes
44
14
9
5
No
3
6
4
1
Didn’t use
34
40
29
24
3. Did you find the computer sessions helpful?
Yes
66
54
31
26
Statistical Computing
Statistical Modelling
Computer Intensive
Nonparametric Smoothing
No
10
3
8
6
Didn’t attend
4
2
3
2
Student costs
The following table summarizes, in aggregate form (£), the invoices received by APTS sending institutions for the four APTS weeks in 2011–12:
Cambridge
Nottingham
Warwick
Glasgow
TOTAL
Registration
fees
£11,160
£9,600
£11,280
£6,000
£38,040
Accommodation
and food
£22,320
£19,490
£21,640
£18,700
£82,150
EPSRC travel
allowance
(£1,001)
(£1,001)
(£0,985)
(£1,004)
(£3,991)
Notes:
Registration fee discounts for students attending all four weeks are all deducted from the final week.
4
APTS: Student registration
1 of 3
http://www2.warwick.ac.uk/fac/sci/statistics/apts/register
Registration for APTS
Student registration for APTS academic year 2012-13 will open on Friday 21st September 2012, and closes at noon on Friday
26th October 2012. Registration applications made after that date will be kept in a priority-ordered reserve list, in case of any
cancellations.
Students can only be registered for APTS weeks by their "sending institution" (i.e., their home department): a list of these
institutions appears below.
If your department wishes to register as a sending institution, then please click here;
If your department wishes to commit to being a full Member Institution of APTS, then please click here.
(All Member Institutions are automatically "sending institutions".)
If your department is included in the list below, the APTS contact (who must be a member of academic staff employed by that
institution) will be provided with a password enabling him/her to complete the student registration form for 2011-12 APTS weeks.
(The student registration form also gives full information on cost.)
The principles and practicalities of student registration and payments include:
date of application within the registration period is unimportant --- it is not used in determining the allocation of APTS
places to students (see the APTS Constitution
for the list of priorities)
sending institutions are invoiced by APTS for the registration fee, and for accommodation/meal costs, of their students
who are allocated APTS places
for EPSRC-funded students taking an APTS week away from home, an allowance is made by APTS to the sending
institution towards the cost of travel
in the case of a student taking all four APTS weeks in the same academic year a 20% rebate of registration fees is made
all financial transactions with individual APTS students, including those relating to travel expenses, are handled by the
sending institution
Please see the FAQ and the Billing and Cancellation policy for more specific information.
List of sending institutions
APTS contact
University of Aberdeen: School of Medical Sciences
Ian Stansfield
Aston University: Mathematics Group
David Saad
University of Bath: Department of Mathematical Sciences
Simon Wood
Queens University Belfast: Department of Sustainability
Frank Figge
University of Birmingham: School of Mathematics
Biman Chakraborty
University of Bristol: Department of Mathematics, Statistics Group
Jonty Rougier
University of Bristol: Department of Social Medicine
Chris Metcalfe
University of Cambridge: CRI
Simon Tavaré
12/09/12 15:02
APTS: Student registration
2 of 3
http://www2.warwick.ac.uk/fac/sci/statistics/apts/register
University of Cambridge: Statistical Laboratory
Susan Pitts
University of Cambridge: MRC Biostatistics Unit
Angela Talbot
Cardiff University: School of Mathematics
Anatoly Zhigljavsky
University College Cork: Statistics Department
David Hawe
Trinity College Dublin: Statistics Group
Simon Wilson
University College Dublin: Statistics Group
Brendan Murphy
Durham University: Department of Mathematical Sciences
Jochen Einbeck
University of East Anglia: Department of Economics
Peter Moffatt
University of East Anglia: Department of Computer Science
Elena Kulinskaya
University of Edinburgh: Roslin Institute
Mark Bronsvoort
University of Edinburgh: School of Mathematics
Natalia Bochkina
University of Essex: Department of Mathematical Sciences
Berthold Lausen
University of Exeter: College of Engineering Mathematics and Physical Sciences
Chris Ferro
National University of Ireland, Galway: Department of Mathematics
John Newell
University of Glasgow: Statistics
Adrian Bowman
Government Communications HQ
Jeremy Bradley
Heriot-Watt University: Department of Actuarial Mathematics and Statistics
George Streftaris
University of Iceland: Department of Mathematics
Gunnar Stefansson
University of Kent: Institute of Mathematics, Statistics and Actuarial Science
Lothar Breuer
Lancaster University: Department of Mathematics and Statistics
Kanchan Mukherjee
Lancaster University: School of Health and Medicine
Peter Diggle
University of Leeds: Department of Statistics
Leonid Bogachev
University of Limerick: Department of Mathematics and Statistics
Gilbert MacKenzie
University of Liverpool: Centre for Medical Statistics and Health Evaluation
Marta García-Fiñana
University of Liverpool: Department of Molecular and Clinical Cancer Medicine
Trevor Cox
University of Liverpool: Department of Mathematical Sciences
Damian Clancy
London School of Hygiene and Tropical Medicine
James Carpenter
King's College London: Department of Statistical Science
Andrew Pickles
University College London: Department of Statistical Science
Afzal Siddiqui
University College London: Department of Infection and Population Health
Andrew Copas
University College London: Institute of Child Health
Mario Cortina-Borja
University of Manchester: School of Mathematics
Peter Foster
National University of Ireland, Maynooth: Department of Mathematics
Caroline Brophy
Newcastle University: Department of Mathematics & Statistics
Colin Gillespie
University of Nottingham: School of Mathematical Sciences
Chris Brignell
Open University: Department of Mathematics & Statistics
Kevin McConway
University of Oxford: Department of Physics
Nick Jones
University of Oxford: Department of Statistics
Jonathan Marchini
University of Oxford: Wellcome Trust Centre for Human Genetics
Gil McVean
University of Plymouth: School of Computing and Mathematics
Julian Stander
University of Reading: Quantitative Biology and Applied Statistics
Fazil Baksh
University of St Andrews: School of Mathematics and Statistics
Carl Donovan
University of Salford: Centre for OR and Applied Statistics
Phil Scarf
12/09/12 15:02
APTS: Student registration
3 of 3
http://www2.warwick.ac.uk/fac/sci/statistics/apts/register
University of Sheffield: Department of Probability and Statistics
Caitlin Buck
University of Sheffield: School of Health and Related Research
Stephen Walters
University of Southampton: School of Mathematics
Robin Mitra
University of Southampton: School of Social Sciences
Peter Smith
University of Surrey: Department of Mathematics
Janet Godolphin
University of Warwick: Complexity Science DTC
Stefan Grosskinsky
University of Warwick: Department of Statistics
David Hobson
University of Warwick: MASDOC Doctoral Training Centre
Christoph Ortner
University of Warwick: Medical School
Nigel Stallard
University of Warwick: Systems Biology Doctoral Training Centre
Vicky Buchanan-Wollaston
University of Wolverhampton: School of Technology
Diwei Zhou
University of York: Department of Mathematics
Stephen Connor
• Contact APTS •
Intranet
Contact APTS
Page contact: David Firth
Last revised: Wed 12 Sep 2012
12/09/12 15:02
APTS: Billing and cancellation policies
1 of 1
http://www2.warwick.ac.uk/fac/sci/statistics/apts/register/b...
Billing and cancellation
This page gives details of the way in which the accounts of sending institutions will be handled, and of the APTS cancellation
policy.
Billing
APTS will maintain an account for each sending institution. Charges made against this account will be:
registration fee for all students
cost of the specified accommodation and food requirements
For students who participate in all four APTS weeks in the same academic year, 20% of registration fees will be rebated.
Invoices will be issued to sending institutions at the end of each APTS week, for the amounts relating to participation in that
APTS week. Registration rebates for students attending all four weeks are made on the invoice for APTS week 4.
Cancellation policy
A statement of this policy will appear also on the form that is signed by sending institutions, to confirm their accepted registrations
for APTS, immediately after the closing date for registering students.
1. Registration fees are payable for all students accepted for an APTS week, and are not normally refunded in the event of
cancellation.
2. In the event of cancellation of a student's participation in an APTS week, the charges made for accommodation and food
will be reduced by
100% if the cancellation is received before noon of the Monday six weeks prior to the Monday of APTS week
50% if the cancellation is received after that but before noon of the Monday four weeks prior to the Monday of
APTS week.
(For an APTS week starting on Tuesday or Wednesday, "the Monday of APTS week" means the preceding Monday.) After
four weeks prior to an APTS week, charges relating to that APTS week are not normally refunded.
Notice of any cancellation should be sent (by the APTS Academic Contact for the student's home department, NOT by the
student concerned) by email to admin@apts.ac.uk.
Intranet
Contact APTS
Page contact: David Firth
Last revised: Thu 22 Sep 2011
12/09/12 09:45
Paper for 4.2 — The Scientific Method, Data Collection and
Quality Control
The Scientific Method, Data Collection and Quality Control
The science of statistics is more than just developing mathematical tools to model data. Inference is
(usually) based on either deductive logic [Pr(D|H)] or inductive logic [Pr(H|D)] – see Gauch (2003). In
an important paper, Box (1976) illustrated how statistical science provided the catalyst for scientists
to iterate between these two inferential states. Davis (2006) provided a modification, emphasising
that this process needed to converge to a point where analysis and inference together create an
action to force a change to the current state.
Of course, at the heart of our subject is data. The moment when an observation is transformed into
data must have integrity, and this part of our subject is, arguably, neglected (Heinzen and Nolan,
2012). It is my view that if we want to be statistical scientists, we should be close to the collection of
the data upon which we choose to base our inference. Chapter 4 in the recent book by Cox and
Donnelly (2011) provides a useful discussion of the issues.
Tim Davis
September 11, 2012
Box, GEP. "Science and statistics". Journal of the American Statistical Association, Vol. 71, No 356,
1976, pp 791-799.
Cox, DR, and Donnelly CA. Principles of applied statistics, CUP, 2011.
Davis, TP. “Science, engineering, and statistics”. Applied Stochastic Models in Business and Industry,
Vol. 22, Issue 5-6, pp 401-430.
Gauch, HG. Jr. Scientific method in practice. CUP, 2003.
Heinzen, TE & Nolan,S. Letter to the editor, RSS News, September 2012.
12
Paper for 4.3 — Statistical Inference: Proposed Syllabus
Received from Jonathan Rougier:
I am attaching a syllabus which I hope you can knock into shape and
submit for the thoughts of the committee. I appreciate that they
could probably do with a lot more information. So perhaps you could
also inform the Advisory Committee that:
1. I will be working with a finite sample space, for simplicity (and
realism!). The parameter space will in general be a convex subset of
R^p.
2. The first lectures will cover aleatory vs epistemic
interpretations of probability (eg history according to Hacking), why
it is that the statistical model might be the same in the two cases
(random sampling and exchangeability), and why it is that the numbers
that emerge are often similar (asymptotic argument and KL divergence).
3. I will not be proposing to interpret Bayes Factors on the Jeffreys/
Good scale, but rather seeing them more as a pragmatic device that
may obviate the need for a careful assessment of the prior
probabilities of the hypotheses. Hence the emphasis on bounds, which
may also obviate the need for a careful assessment of Pr(theta |
H_1). This material goes back to Edwards, Lindman and Savage; it has
been picked up in medical statistics by Goodman.
4. I’ll prove that P-values are sub-uniform under H_0 in a handout (I
have not seen a proof of this, but it is true!). In the lecture I’ll
just prove that they are uniform for continuous X and simple H_0.
This is enough to understand why small P-values may weaken H_0 but
large values cannot support it. I’ll use David Cox’s construction
(from Savage et al, 1962) to link significance tests and UMP
hypothesis tests via the one-parameter exponential family. This then
ties back to bounds on Bayes factors in a very neat way.
5. No sufficiency (too specific for a general inference course).
Hence a slightly stronger Weak Conditionality Principle is used to
achieve equivalence with the Likelihood Principle. LP implies that
one should condition on experimental ancillary values (eg X in
regression, random N, missing at random data), the stopping principle
will be explained through ’significance hunting’, also known as
’researcher degrees of freedom’.
13
The proposed syllabus is:
Statistical Inference
Module leader: J.C. Rougier
Aims: To examine different statistical approaches for evaluating the evidential support for or against
hypotheses, both the tools (e.g. P -values, Bayes factors), and the principles.
Learning outcomes: Familiarity with different approaches to statistical reasoning, and a clear understanding of what can and cannot be inferred from each approach. After taking the module, students
should be able to recognise and avoid the common inferential fallacies: the Prosecutor’s fallacy, the base
rate fallacy, and the P -value fallacy.
Prerequisites: Preliminary reading material will cover: the probability calculus, conditional probability,
Bayes’s theorem, convergence in probability and the Weak Law of Large Numbers; maximum likelihood
(ML) estimation, confidence sets; significance tests (i.e. tests of goodness of fit), hypothesis tests, the
Neyman-Pearson Lemma, uniformly most powerful (UMP) tests; Bayesian estimation, prediction, and
hypothesis testing. Some of these will be revisited over the course of the module. Familiarity with R will
be useful, for simple Monte Carlo simulations.
Further information on all of these topics can be found in a standard undergraduate statistics textbook, for example
• J.A. Rice, 1999, Mathematical Statistics and Data Analysis, 2nd edn, Duxbury Press (more recent
edition available); or
• Morris H, DeGroot, and Mark J Schervish, 2002, Probability & Statistics, Addison Wesley, 3rd edn.
Topics:
1. Frequentist and Bayesian approaches to inference.
2. Bayes factors, and their approximation and bounds.
3. P -values, various inferential fallacies.
4. Conditionality and the Likelihood Principle (LP).
5. Implications of the LP, ancillarity, stopping rules.
Assessment: Exam-style questions on the implementation of different approaches in particular types of
inference, possibly involving additional reading.
14
The current syllabus, for reference is:
Statistical Inference
Module leader: D. Firth
Aims: This module will provide students with a solid understanding of the main approaches to statistical
inference, their strengths and limitations, their similarities and differences, and their role in underpinning
statistical methodology.
Learning outcomes: After taking this module students should have an appreciation of the predominant
modes of inference and their inter-relationships, and should be better equipped to read the published
literature on both technical and foundational aspects of inference.
Prerequisites: Students should review the following definitions and results: likelihood, sufficiency, Bayes’
theorem; simple properties of normal, exponential, binomial and Poisson distributions; linear model and
the method of least squares.
Topics:
• Role of formal inference, nature of probability, frequentist and Bayesian approaches.
• Role of sufficiency; role of Neyman-Pearson theory; relation between significance tests and confidence limits.
• Maximum likelihood and associated issues; properties in ’standard’ situations, and in some more
difficult cases.
• Exponential-family models.
• Other approaches (e.g., estimating equations, pseudo-likelihoods).
Assessment one of:
• An essay on one of a list of topics suggested by the module leader.
• Report of a numerical investigation on one of a list of topics suggested by the module leader, e.g.,
comparing Bayesian and frequentist approaches to the analysis of a particular model, or assessing
the accuracy of inferences based on large-sample approximation.
15
Written Contributions
David Elston (BioSS)
Dear Wilf & David,
Please accept my apologies for absence from the APTS Advisory Committee meeting (AC6) on
13th September 2012.
I wish to make four observations.
1) The system as a whole seems to be working well, which is good, but I wonder whether the
cause for the drop-off in attendance through the weeks has been investigated and responded
to (from the questionnaire returns, 84 week 1, 62 week 2, 43 week 3).
2) I think the Asymptotics course is not the best place to deal with some of the issues
raised by high-dimensional data: to my mind, the Inference course is the best place for
some issues, particularly the shift from thinking about false positive rates (exemplified
by p-values) to false discovery rates (having set a p-value threshold to identify ’discoveries’,
what proportion of the putative discoveries are likely to be genuine). Although these methods
are used in a bioinformatics context, they are much more general than that and have a role
in data mining approaches to extracting information from large databases.
3) I can understand not having a distinct module on Design, but I would be both surprised
and disappointed if design issues were not raised within existing modules. If not, then
we are training statisticians to be expert at analysis, but not to be rounded members of
collaborative teams, in which the statistical expert is expected to guide data collection
as well as analysis. The idea that statisticians should be involved in the planning of
designed experiments but not the planning of collection of data in observational studies
is a common mis-understanding which we should strive to dispell.
It would seem natural to be to have sections on design in (at least) the modules on:
Statistical Inference (in association with power calculations, using design as a way of
reducing variation between comparisons of interest);
Statistical Modelling (principles of D-optimality and A optimality as ways of measuring
the information content associated with sets of design points);
Spatial and Longitudinal Data Analysis (eg layout of sampling points to obtain a good spread
of information about the variogram); and
Non-parametric Smoothing (how should expectation about the shape of the response guide
the collection of apporpriate data).
These sections do not have to be large, but they should surely be there!
4) I have advocated previously but without success some kind of ’overview of the discipline
of statistics’ lecture. I still think this would be a good way of putting the modules into
context, and laying out what’s not covered to encourage the cohort of PhD students to see
the APTS courses and their own PhD work as part of a (necessarily) much wider discipline.
I hope the meeting goes well, and look forward to seeing a copy of the minues.
David
-Biomathematics and Statistics Scotland (BioSS) is formally part of The
James Hutton Institute (JHI), a registered Scottish charity No. SC041796
and a company limited by guarantee No. SC374831
16
Jonathan Rougier (Bristol)
Also, some brief comments for the Advisory Committee meeting. I
would strongly support a module on high-dimensional inference,
including high throughput assessment, multiple testing, and BH
screening to control the FDR. I would also support more material on
classification, which is a bridge into machine learning.
and the later addition
And can I just support David Elston’s point 2. If I had had 1.5 more hours I would like
to have covered multiple testing and high dimensional screening in the Statistical Inference.
Working through the proof of the BH FDR bound would be a good experience for the students.
But the underlying theme of my syllabus is "Know what you can and what you cannot infer"
and I think the material on ancillarity and stopping rules in the final slot is central
to current practice, eg in medical science.
Jonty
Dave Woods (Southampton)
Dear Adam and Robin
As Robin will be attending the meeting tomorrow to represent Southampton, I thought I would
comment briefly on the the discussion from both Tim Davis and David Elston on including
Design in APTS.
As discussed at the last executive committee, I am planning on including a short session
(~1 - 1.5 hours) on Design of Experiments in the Stats Modelling module. This is will include
the fundamental principals of design (so at least nodding in the direction of Tim’s comments)
and the basics of how a good design can allow for more informative modelling (cf David’s
comments).
[For the latter topic, my main focus was going to be on confounding and bias (both deliberate
and accidental), as this seems to fit in well with model choice, a main theme of the Stats
Modelling course.]
Thanks and best wishes
Dave
Shuaiwei Zhou (Student Representative; TCD)
As I was given the role as a student representative after the APTS at Cambridge and I did
not attend the APTS at Warwick due to the visa issue,
the only feedbacks I collected are about the APTS in Nottingham and Glasgow.
APTS at Nottingham:
1. The students generally believed that the Asymptotics Statistics was kind of challenge
to understand. Generally speaking (as different students have varied levels), we would prefer
the leader of the module spend more time explaining the basic theories and applications
in Asymptotics and it would be preferable if more exercises could be mentioned and instructed
in class.
2. The students found that the dinners at Nottingham cost much time, which were 2-hours
each day. Actually we have to admit that the dinner at Nottingham were brilliant.
17
APTS at Glasgow:
1. For the Spatial and Longitudinal Data Analysis, we would like to say if the leader of
this module could provide a more detailed lecture notes rather than slides.
2. For the Nonparametric Smoothing, a few students talked to me that they prefer to have
the notes available in advance rather than having to take notes in class.
In all, we really appreciate the organisations in every APTS module, which
have inspired us very much and gave us the entrance to Statistics and a
wonderful chance to meet the fellows in Statistics in UK and beyond. Also Nat
and I created a mail list for the students who registered in the list and a
Facebook page for the fellows, which we believe would enhance the
communication between us.
18